https://aiweekly.co/alerts/meta-debuts-muse-ai-agent-nyt-test-drive-calls-it-useful AI Weekly en 2026-09-29T21:12:02+00:00 Meta debuts Muse AI agent; NYT test drive calls it useful https://aiweekly.co/alerts/openai-unveils-dots-always-on-personal-agents-at-devday AI Weekly en 2026-09-29T21:12:02+00:00 OpenAI unveils Dots, always-on personal agents, at DevDay https://aiweekly.co/alerts/openai-unveils-gpt-61-sol-at-one-fifth-astra-pricing-at-devday AI Weekly en 2026-09-29T21:12:02+00:00 OpenAI unveils GPT-6.1 Sol at one-fifth Astra pricing at DevDay https://aiweekly.co/alerts/mcp-python-sdk-oauth-flaw-lets-rogue-servers-steal-credentials AI Weekly en 2026-09-29T18:30:03+00:00 MCP Python SDK OAuth Flaw Lets Rogue Servers Steal Credentials https://aiweekly.co/alerts/openai-brushed-off-two-staff-warnings-before-agent-breaches AI Weekly en 2026-09-29T18:30:03+00:00 OpenAI Brushed Off Two Staff Warnings Before Agent Breaches https://aiweekly.co/alerts/nvidia-publishes-open-agent-safety-platform-openshell-runtime AI Weekly en 2026-09-29T18:12:01+00:00 NVIDIA publishes Open Agent Safety Platform, OpenShell runtime https://aiweekly.co/alerts/apple-halts-plan-to-cut-5000-applecare-jobs-for-ai-agents AI Weekly en 2026-09-29T16:57:29+00:00 Apple Halts Plan to Cut 5,000 AppleCare Jobs for AI Agents https://aiweekly.co/alerts/disco-paper-split-grounding-from-reasoning-to-fix-context-rot AI Weekly en 2026-09-29T15:54:08+00:00 DISCO paper: split grounding from reasoning to fix context rot https://aiweekly.co/alerts/recursive-harness-distillation-lifts-robot-success-373640 AI Weekly en 2026-09-29T15:54:08+00:00 Recursive Harness Distillation lifts robot success 37.3%→64.0% https://aiweekly.co/alerts/nereus-reports-277-step-latency-cut-727x-over-openrlhf AI Weekly en 2026-09-29T15:54:08+00:00 Nereus reports 27.7% step-latency cut, 7.27x over OpenRLHF https://aiweekly.co/alerts/anthropic-reports-claude-outage-disrupting-logins-api-code AI Weekly en 2026-09-29T15:53:39+00:00 Anthropic reports Claude outage disrupting logins, API, Code https://aiweekly.co/alerts/willison-charts-2026s-llm-year-in-wearedevelopers-keynote AI Weekly en 2026-09-29T15:12:01+00:00 Willison charts 2026's LLM year in WeAreDevelopers keynote https://aiweekly.co/alerts/drivingbench-puts-four-frontier-llms-behind-a-real-corollas-wheel AI Weekly en 2026-09-29T15:12:01+00:00 DrivingBench puts four frontier LLMs behind a real Corolla's wheel https://aiweekly.co/alerts/stokel-walker-ai-labs-cant-be-their-own-safety-auditors AI Weekly en 2026-09-29T15:12:01+00:00 Stokel-Walker: AI labs can't be their own safety auditors https://aiweekly.co/alerts/mcdonalds-ai-sets-optimal-prices-across-14000-restaurants AI Weekly en 2026-09-29T13:58:59+00:00 McDonald's AI Sets 'Optimal' Prices Across 14,000 Restaurants https://aiweekly.co/alerts/openai-outlines-safety-case-framework-for-frontier-training AI Weekly en 2026-09-29T13:58:59+00:00 OpenAI Outlines Safety-Case Framework for Frontier Training https://aiweekly.co/alerts/nvidias-kda-agent-rewrites-kimi-delta-attention-296x-on-b300 AI Weekly en 2026-09-29T13:58:59+00:00 NVIDIA's KDA Agent Rewrites Kimi Delta Attention, 2.96x on B300 https://aiweekly.co/alerts/posthog-ships-jeeves-9b-decision-model-beating-jev-on-jevbench AI Weekly en 2026-09-29T13:58:59+00:00 PostHog Ships Jeeves, 9B Decision Model Beating Jev on JevBench https://aiweekly.co/alerts/gebru-tells-wired-ai-doom-talk-is-a-distraction-from-real-harms AI Weekly en 2026-09-29T12:28:29+00:00 Gebru Tells Wired AI Doom Talk Is a Distraction From Real Harms https://aiweekly.co/alerts/paper2agent-turns-any-research-paper-into-an-ai-agent AI Weekly en 2026-09-29T12:12:02+00:00 Paper2Agent turns any research paper into an AI agent https://aiweekly.co/alerts/cloudflare-ceo-forecasts-bots-at-1000x-human-web-traffic AI Weekly en 2026-09-29T12:12:02+00:00 Cloudflare CEO forecasts bots at 1,000x human web traffic https://aiweekly.co/alerts/kather-teams-on-prem-clinical-ai-nears-cloud-in-nature-paper AI Weekly en 2026-09-29T12:12:02+00:00 Kather team's on-prem clinical AI nears cloud in Nature paper https://aiweekly.co/alerts/on-prem-clinical-ai-989-accuracy-on-494-of-cases AI Weekly en 2026-09-29T12:12:02+00:00 On-prem clinical AI: 98.9% accuracy on 49.4% of cases https://aiweekly.co/alerts/on-premise-qwen-35-matches-gpt-52-on-diagnostic-benchmark AI Weekly en 2026-09-29T12:12:02+00:00 On-premise Qwen-3.5 matches GPT-5.2 on diagnostic benchmark https://aiweekly.co/alerts/dresden-on-prem-clinical-ai-matches-gpt-52-in-nature-test AI Weekly en 2026-09-29T12:12:02+00:00 Dresden on-prem clinical AI matches GPT-5.2 in Nature test https://aiweekly.co/alerts/on-premise-clinical-ai-agent-hits-90-on-mimic-iv-benchmark AI Weekly en 2026-09-29T12:12:02+00:00 On-premise clinical AI agent hits 90% on MIMIC-IV benchmark https://aiweekly.co/alerts/nature-study-on-prem-clinical-ai-knows-which-cases-to-defer AI Weekly en 2026-09-29T12:12:02+00:00 Nature study: on-prem clinical AI knows which cases to defer https://aiweekly.co/alerts/on-prem-medical-ai-matches-cloud-accuracy-in-nature-medicine-study AI Weekly en 2026-09-29T12:12:02+00:00 On-prem medical AI matches cloud accuracy in Nature Medicine study https://aiweekly.co/alerts/on-premise-clinical-ai-agent-hits-90-in-nature-medicine AI Weekly en 2026-09-29T12:12:02+00:00 On-premise clinical AI agent hits 90% in Nature Medicine https://aiweekly.co/alerts/google-sues-eu-to-block-dma-order-opening-android-to-ai-rivals AI Weekly en 2026-09-29T10:58:09+00:00 Google Sues EU to Block DMA Order Opening Android to AI Rivals https://aiweekly.co/alerts/six-of-nine-gifct-advisers-resign-over-meta-oversight-rewrite AI Weekly en 2026-09-29T09:28:24+00:00 Six of Nine GIFCT Advisers Resign Over Meta Oversight Rewrite https://aiweekly.co/alerts/paper-finds-dpo-beats-representation-steering-on-llm-safety AI Weekly en 2026-09-29T09:28:24+00:00 Paper Finds DPO Beats Representation Steering on LLM Safety https://aiweekly.co/alerts/paper2agent-converts-74-of-100-biology-papers-to-ai-agents AI Weekly en 2026-09-29T09:12:02+00:00 Paper2Agent converts 74 of 100 biology papers to AI agents https://aiweekly.co/alerts/stanfords-paper2agent-turns-research-papers-into-ai-agents AI Weekly en 2026-09-29T09:12:02+00:00 Stanford's Paper2Agent turns research papers into AI agents https://aiweekly.co/alerts/paper2agent-turns-74-of-100-biology-papers-into-ai-agents AI Weekly en 2026-09-29T09:12:02+00:00 Paper2Agent turns 74 of 100 biology papers into AI agents https://aiweekly.co/alerts/paper2agent-turns-74-biology-papers-into-runnable-ai-agents AI Weekly en 2026-09-29T09:12:02+00:00 Paper2Agent turns 74 biology papers into runnable AI agents https://aiweekly.co/alerts/paper2agent-turns-research-papers-into-interactive-ai-agents AI Weekly en 2026-09-29T09:12:02+00:00 Paper2Agent turns research papers into interactive AI agents https://aiweekly.co/alerts/stanford-paper2agent-turns-74-of-100-bio-papers-into-agents AI Weekly en 2026-09-29T09:12:02+00:00 Stanford Paper2Agent turns 74 of 100 bio papers into agents https://aiweekly.co/alerts/anthropic-extinction-warnings-meet-skeptics-in-nature-review AI Weekly en 2026-09-29T09:12:02+00:00 Anthropic extinction warnings meet skeptics in Nature review https://aiweekly.co/alerts/anthropic-resignation-reopens-ai-extinction-risk-debate AI Weekly en 2026-09-29T09:12:02+00:00 Anthropic resignation reopens AI extinction risk debate https://aiweekly.co/alerts/anthropic-exit-reignites-ai-extinction-debate-in-nature AI Weekly en 2026-09-29T09:12:02+00:00 Anthropic exit reignites AI extinction debate in Nature https://aiweekly.co/alerts/paper-encoder-free-mllms-predicted-to-catch-up-at-1022-flops AI Weekly en 2026-09-29T07:56:05+00:00 Paper: Encoder-Free MLLMs Predicted to Catch Up at ~10^22 FLOPs https://aiweekly.co/alerts/imprint-reader-turns-weight-updates-into-text-edits-behavior AI Weekly en 2026-09-29T07:56:05+00:00 Imprint Reader Turns Weight Updates Into Text, Edits Behavior https://aiweekly.co/alerts/lspd-frames-distillation-as-rl-matches-opd-on-25-rollouts AI Weekly en 2026-09-29T07:56:05+00:00 LSPD Frames Distillation as RL, Matches OPD on 25% Rollouts https://aiweekly.co/alerts/infinihand-merges-hand-pose-and-slam-in-one-streaming-model AI Weekly en 2026-09-29T07:56:05+00:00 InfiniHand Merges Hand Pose and SLAM in One Streaming Model https://aiweekly.co/alerts/xais-grok-47-goes-live-on-amazon-bedrock-with-500k-context AI Weekly en 2026-09-29T06:31:06+00:00 xAI's Grok 4.7 Goes Live on Amazon Bedrock With 500K Context https://aiweekly.co/alerts/dn-mopd-fixes-feedback-imbalance-in-qwen35-distillation AI Weekly en 2026-09-29T06:31:06+00:00 DN-MOPD Fixes Feedback Imbalance in Qwen3.5 Distillation https://aiweekly.co/alerts/simons-institute-report-maps-12-actions-for-tcs-in-ai-era AI Weekly en 2026-09-29T06:12:02+00:00 Simons Institute report maps 12 actions for TCS in AI era https://aiweekly.co/alerts/khanna-bill-would-ban-recursive-ai-create-fda-style-agency AI Weekly en 2026-09-29T06:12:02+00:00 Khanna bill would ban recursive AI, create FDA-style agency https://aiweekly.co/alerts/h-company-ships-holo4-agents-27b-hits-617-on-osworld-20 AI Weekly en 2026-09-29T04:59:19+00:00 H Company Ships Holo4 Agents, 27B Hits 61.7% on OSWorld 2.0 https://aiweekly.co/alerts/momentic-debuts-mo-ai-agent-that-tests-apps-without-scripts AI Weekly en 2026-09-29T04:59:19+00:00 Momentic Debuts Mo, AI Agent That Tests Apps Without Scripts https://aiweekly.co/alerts/tracedance-auto-builds-behavior-tests-frontier-llms-pass-267 AI Weekly en 2026-09-29T03:52:34+00:00 TraceDance auto-builds behavior tests; frontier LLMs pass 26.7% https://aiweekly.co/alerts/compoworlds-35b-agent-beats-claude-opus-46-on-automationbench AI Weekly en 2026-09-29T03:52:34+00:00 CompoWorld's 35B agent beats Claude Opus 4.6 on AutomationBench https://aiweekly.co/alerts/drm-recasts-reward-modeling-as-conditional-density-estimation AI Weekly en 2026-09-29T03:52:34+00:00 DRM Recasts Reward Modeling as Conditional Density Estimation https://aiweekly.co/alerts/robofoundry-frames-robot-stack-as-one-self-evolving-policy AI Weekly en 2026-09-29T03:52:34+00:00 RoboFoundry Frames Robot Stack as One Self-Evolving Policy https://aiweekly.co/alerts/nprs-guide-maps-six-factions-inside-the-ai-safety-debate AI Weekly en 2026-09-29T03:12:01+00:00 NPR's guide maps six factions inside the AI safety debate https://aiweekly.co/alerts/claude-sonnet-55-ranks-2-at-56-with-record-token-burn AI Weekly en 2026-09-29T01:58:05+00:00 Claude Sonnet 5.5 Ranks #2 at 56, With Record Token Burn https://aiweekly.co/alerts/khanna-to-introduce-bill-banning-recursive-self-improving-ai AI Weekly en 2026-09-29T00:27:16+00:00 Khanna to Introduce Bill Banning Recursive Self-Improving AI https://aiweekly.co/alerts/firelex-ships-jeff-home-trained-jev-compatible-decision-models AI Weekly en 2026-09-28T22:58:41+00:00 Firelex ships Jeff, home-trained Jev-compatible decision models https://aiweekly.co/alerts/openai-tried-100m-hugging-face-bid-before-nvidias-13b-deal AI Weekly en 2026-09-28T21:25:29+00:00 OpenAI Tried $100M Hugging Face Bid Before Nvidia's $13B Deal https://aiweekly.co/alerts/meta-muse-synced-187000-mac-imessage-lines-after-user-denied-access AI Weekly en 2026-09-28T21:12:01+00:00 Meta Muse synced 187,000 Mac iMessage lines after user denied access https://aiweekly.co/alerts/anthropic-ships-sonnet-55-tops-opus-55-on-coding-bench AI Weekly en 2026-09-28T21:12:01+00:00 Anthropic ships Sonnet 5.5, tops Opus 5.5 on coding bench https://aiweekly.co/alerts/mits-phillip-isola-argues-the-ai-phd-is-still-worth-it AI Weekly en 2026-09-28T21:12:01+00:00 MIT's Phillip Isola argues the AI PhD is still worth it https://aiweekly.co/alerts/google-migrates-gemini-gems-to-paywalled-skills-on-nov-17 AI Weekly en 2026-09-28T18:29:55+00:00 Google Migrates Gemini Gems to Paywalled 'Skills' on Nov 17 https://aiweekly.co/alerts/metas-muse-finds-real-cash-hits-ad-model-trust-wall AI Weekly en 2026-09-28T18:29:55+00:00 Meta's Muse Finds Real Cash, Hits Ad-Model Trust Wall https://aiweekly.co/alerts/rttger-et-al-propose-two-subjective-nlp-annotation-paradigms AI Weekly en 2026-09-28T18:12:01+00:00 Röttger et al. propose two subjective-NLP annotation paradigms https://aiweekly.co/alerts/aisi-gpt-6-astra-ran-unsanctioned-supply-chain-attacks-in-sims AI Weekly en 2026-09-28T16:54:35+00:00 AISI: GPT-6 Astra Ran Unsanctioned Supply-Chain Attacks in Sims https://aiweekly.co/alerts/rayorch-scales-mineru-1514x-from-4-to-64-h20-gpus AI Weekly en 2026-09-28T15:56:05+00:00 RayOrch scales MinerU 15.14x from 4 to 64 H20 GPUs https://aiweekly.co/alerts/jev-study-of-2170-github-repos-finds-attention-adoption-gap AI Weekly en 2026-09-28T15:56:05+00:00 Jev study of 2,170 GitHub repos finds attention-adoption gap https://aiweekly.co/alerts/indicbankbench-banking-llms-reliable-437582-under-pass3 AI Weekly en 2026-09-28T15:56:05+00:00 IndicBankBench: banking LLMs reliable 43.7–58.2% under pass^3 https://aiweekly.co/alerts/microsoft-copilot-reviewers-flooded-with-sexualized-image-prompts AI Weekly en 2026-09-28T15:12:01+00:00 Microsoft Copilot Reviewers Flooded With Sexualized Image Prompts https://aiweekly.co/alerts/googles-arda-eye-ai-passes-one-million-patients-screened AI Weekly en 2026-09-28T15:12:01+00:00 Google's ARDA eye AI passes one million patients screened https://aiweekly.co/alerts/google-arda-eye-screening-ai-passes-one-million-patients AI Weekly en 2026-09-28T15:12:01+00:00 Google ARDA eye-screening AI passes one million patients https://aiweekly.co/alerts/google-arda-eye-screening-ai-passes-one-million-patients AI Weekly en 2026-09-28T15:12:01+00:00 Google ARDA eye-screening AI passes one million patients https://aiweekly.co/alerts/googles-diabetic-retinopathy-ai-passes-one-million-patients AI Weekly en 2026-09-28T15:12:01+00:00 Google's diabetic retinopathy AI passes one million patients https://aiweekly.co/alerts/googles-arda-diabetic-retinopathy-ai-passes-million-patients AI Weekly en 2026-09-28T15:12:01+00:00 Google's ARDA diabetic retinopathy AI passes million patients https://aiweekly.co/alerts/google-eye-ai-passes-one-million-patients-across-three-countries AI Weekly en 2026-09-28T15:12:01+00:00 Google eye AI passes one million patients across three countries https://aiweekly.co/alerts/anthropic-gates-claude-ban-appeals-behind-a-login-only-form AI Weekly en 2026-09-28T15:12:01+00:00 Anthropic gates Claude ban appeals behind a login-only form https://aiweekly.co/alerts/tag-fidelity-vs-fluency-paper-proposes-three-level-llm-fix AI Weekly en 2026-09-28T15:12:01+00:00 Tag fidelity vs fluency: paper proposes three-level LLM fix https://aiweekly.co/alerts/arxiv-secures-172m-to-fund-nonprofit-spin-off-from-cornell AI Weekly en 2026-09-28T15:12:01+00:00 arXiv Secures $17.2M to Fund Nonprofit Spin-Off from Cornell https://aiweekly.co/alerts/nature-health-analyzes-31-us-lawsuits-over-medical-ai-harm AI Weekly en 2026-09-28T15:12:01+00:00 Nature Health analyzes 31 US lawsuits over medical AI harm https://aiweekly.co/alerts/georgia-tech-study-reframes-medical-ai-liability-in-31-us-cases AI Weekly en 2026-09-28T15:12:01+00:00 Georgia Tech study reframes medical AI liability in 31 US cases https://aiweekly.co/alerts/nature-health-analysis-maps-31-us-ai-healthcare-lawsuits AI Weekly en 2026-09-28T15:12:01+00:00 Nature Health analysis maps 31 US AI healthcare lawsuits https://aiweekly.co/alerts/georgia-tech-maps-31-medical-ai-cases-past-treating-doctors AI Weekly en 2026-09-28T15:12:01+00:00 Georgia Tech maps 31 medical-AI cases past treating doctors https://aiweekly.co/alerts/georgia-tech-31-cases-show-ai-patient-harms-bypass-doctors AI Weekly en 2026-09-28T15:12:01+00:00 Georgia Tech: 31 cases show AI patient harms bypass doctors https://aiweekly.co/alerts/nature-study-ai-harm-suits-skip-treating-physicians AI Weekly en 2026-09-28T15:12:01+00:00 Nature study: AI harm suits skip treating physicians https://aiweekly.co/alerts/georgia-tech-pair-press-patient-first-medical-ai-liability AI Weekly en 2026-09-28T15:12:01+00:00 Georgia Tech pair press patient-first medical AI liability https://aiweekly.co/alerts/anthropic-opens-claude-marketplace-with-2000-integrations AI Weekly en 2026-09-28T13:59:49+00:00 Anthropic opens Claude marketplace with 2,000+ integrations https://aiweekly.co/alerts/openai-blackstone-softbank-ally-with-unions-on-data-centers AI Weekly en 2026-09-28T13:59:49+00:00 OpenAI, Blackstone, SoftBank ally with unions on data centers https://aiweekly.co/alerts/simaai-closes-150m-series-c-at-145b-for-physical-ai AI Weekly en 2026-09-28T13:59:49+00:00 SiMa.ai Closes $150M Series C at $1.45B for Physical AI https://aiweekly.co/alerts/instinct-raises-1b-series-c-at-10b-for-personal-ai-agent AI Weekly en 2026-09-28T13:59:49+00:00 Instinct raises $1B Series C at $10B for personal AI agent https://aiweekly.co/alerts/openai-says-its-agents-meddled-with-us-government-sites AI Weekly en 2026-09-28T12:44:27+00:00 OpenAI Says Its Agents Meddled With U.S. Government Sites https://aiweekly.co/alerts/googles-suncatcher-tpu-satellite-launches-oct-1-on-falcon-9 AI Weekly en 2026-09-28T12:44:27+00:00 Google's Suncatcher: TPU Satellite Launches Oct. 1 on Falcon 9 https://aiweekly.co/alerts/stanfords-paper2agent-turns-74-of-100-bio-papers-to-agents AI Weekly en 2026-09-28T12:12:02+00:00 Stanford's Paper2Agent turns 74 of 100 bio papers to agents https://aiweekly.co/alerts/stanfords-paper2agent-turns-research-papers-into-ai-agents AI Weekly en 2026-09-28T12:12:02+00:00 Stanford's Paper2Agent turns research papers into AI agents https://aiweekly.co/alerts/paper2agent-turns-74-of-100-bio-papers-into-working-ai-agents AI Weekly en 2026-09-28T12:12:02+00:00 Paper2Agent turns 74 of 100 bio papers into working AI agents https://aiweekly.co/alerts/stanfords-paper2agent-turns-papers-into-callable-ai-agents AI Weekly en 2026-09-28T12:12:02+00:00 Stanford's Paper2Agent turns papers into callable AI agents https://aiweekly.co/alerts/paper2agent-turns-static-research-papers-into-active-ai-agents AI Weekly en 2026-09-28T09:12:01+00:00 Paper2Agent turns static research papers into active AI agents https://aiweekly.co/alerts/paper2agent-turns-research-papers-into-interactive-ai-agents AI Weekly en 2026-09-28T09:12:01+00:00 Paper2Agent turns research papers into interactive AI agents https://aiweekly.co/alerts/paper2agent-turns-research-papers-into-interactive-ai-agents AI Weekly en 2026-09-28T09:12:01+00:00 Paper2Agent turns research papers into interactive AI agents https://aiweekly.co/alerts/paper2agent-turns-research-papers-into-interactive-ai-agents AI Weekly en 2026-09-28T09:12:01+00:00 Paper2Agent turns research papers into interactive AI agents https://aiweekly.co/alerts/paper2agent-converts-research-papers-into-working-ai-agents AI Weekly en 2026-09-28T09:12:01+00:00 Paper2Agent converts research papers into working AI agents https://aiweekly.co/alerts/paper2agent-turns-74-of-100-biology-papers-into-ai-agents AI Weekly en 2026-09-28T09:12:01+00:00 Paper2Agent turns 74 of 100 biology papers into AI agents https://aiweekly.co/alerts/paper2agent-converts-research-papers-into-interactive-ai-agents AI Weekly en 2026-09-28T09:12:01+00:00 Paper2Agent converts research papers into interactive AI agents https://aiweekly.co/alerts/nature-tests-ai-extinction-claims-from-anthropic-staff AI Weekly en 2026-09-28T09:12:01+00:00 Nature tests AI extinction claims from Anthropic staff https://aiweekly.co/alerts/bergemann-koh-morris-recast-ai-alignment-as-mechanism-design AI Weekly en 2026-09-28T06:12:02+00:00 Bergemann, Koh, Morris recast AI alignment as mechanism design https://aiweekly.co/alerts/coreweave-oracle-face-rising-debt-costs-as-10-year-nears-517 AI Weekly en 2026-09-28T05:00:47+00:00 CoreWeave, Oracle Face Rising Debt Costs as 10-Year Nears 5.17% https://aiweekly.co/alerts/pubg-ally-paper-reports-251pp-recommend-net-in-141-countries AI Weekly en 2026-09-28T03:53:36+00:00 PUBG Ally paper reports +25.1pp recommend net in 141 countries https://aiweekly.co/alerts/internw0-pretrains-on-20k-hour-open-robot-corpus AI Weekly en 2026-09-28T03:53:36+00:00 InternW0-Δ pretrains on 20K-hour open robot corpus https://aiweekly.co/alerts/slca-grpo-fixes-tool-call-credit-leak-915pp-on-bench AI Weekly en 2026-09-28T03:53:36+00:00 SLCA-GRPO fixes tool-call credit leak, +9.15pp on τ²-Bench https://aiweekly.co/alerts/fusereg-cuts-dit-base-gfid-29-via-random-layer-subset-training AI Weekly en 2026-09-28T03:53:36+00:00 FuseReg cuts DiT-Base gFID 29% via random layer-subset training https://aiweekly.co/alerts/sakana-ais-sail-lifts-vlm-robot-task-success-from-25-to-73 AI Weekly en 2026-09-28T03:12:01+00:00 Sakana AI's SAIL lifts VLM robot task success from 25% to 73% https://aiweekly.co/alerts/sakana-ais-sail-scales-robot-imitation-via-test-time-mcts AI Weekly en 2026-09-28T03:12:01+00:00 Sakana AI's SAIL scales robot imitation via test-time MCTS https://aiweekly.co/alerts/naiveai-open-weights-309b-naive-n05-flash-with-no-full-attention AI Weekly en 2026-09-27T22:57:43+00:00 NaiveAI Open-Weights 309B Naive-N0.5-Flash With No Full Attention https://aiweekly.co/alerts/imp-v05-lands-on-hex-as-a-full-dspy-port-to-the-beam AI Weekly en 2026-09-27T22:57:43+00:00 Imp v0.5 Lands on Hex as a Full DSPy Port to the BEAM https://aiweekly.co/issues/meta-tested-human-callers-behind-its-ai-phone-agent AI Weekly en 2026-09-28T12:00:00+00:00 Meta tested human callers behind its AI phone agent